This repository is a Python implementation for GMM-HMM model from scratch using Viterbi method. Also, we would use this model for recognizing letters from the sequence of the writing movements. The dataset is place in the data folder of the repository which includes writing sequence of 5 letters of a, e, i, o, u.
Clone the program.
git clone https://github.com/raminnakhli/GMM-HMM-from-scratch.git
The requirements are some common packages in machine learning. You can install them using below command.
pip install -r requirement.txt
Now, you can run the experiments with default configuration using the below command.
python main.py
In addition, one can change configuration of the program using command-line flags while running the above command, which are explained in the following section.
You can change configuration of the model using the below flags.
| Short Format | Long Format | Valid Values | Explanation |
|---|---|---|---|
| -ht | --hyperparameter-test | No Value | sets random values for state and mixture count, runs training for 5 times, and reports the best parameter |
| -stpr Stop_Rate | --stop-rate Stop_Rate | Float Value | specifies the stop difference criteria of the EM algorithm in GMM-HMM model |
| -stc State_Count | --state-count State_Count | Integer Value | specifies the number of Markov states |
| -mc Mitxure_Count | --mixture-count Mitxure_Count | Integer Value | specifies the number of mixture in GMM |
| -blk | --belkin | No Value | enables using Belkin method in GMM-HMM training |
| -vtb | --viterbi | No Value | enabled using Viterbi method in GMM-HMM training |
| -vft | --viterbi-forward-test | No Value | enabled comparison between Viterbi and forward accuracy |
Contributions are what make the open source community such an amazing place to be learn, inspire, and create. Any contributions you make are greatly appreciated.
- Fork the Project
- Create your Feature Branch (
git checkout -b feature/AmazingFeature) - Commit your Changes (
git commit -m 'Add some AmazingFeature') - Push to the Branch (
git push origin feature/AmazingFeature) - Open a Pull Request
Ramin Ebrahim Nakhli - raminnakhli@gmail.com
Project Link: https://github.com/raminnakhli/GMM-HMM-from-scratch